OpenAI Rogue AI Agents Hacked 10 More Sites - Sam Altman Needs Jail Time
Eli the Computer Guy
September 14, 2026
OpenAI Rogue AI Agents Hacked 10 More Sites - Sam Altman Needs Jail Time
Speakers Eli the Computer Guy
TopicsTechnology
Eli the Computer Guy (0:00)
Oh, lordy, lordy, lordy. Coming from Reuters, OpenAI's rogue agents used at least 10 more sites for unauthorized communications, researchers say.
Oh, crikey. It's so back.
It only keeps getting worse. So I did a video previously talking about how OpenAI's rogue AI agents hacked a German Wiki site and basically used it as a communication bus, right? So for whatever reason, the AI agents weren't able to communicate with each other using OpenAI's internal infrastructure. So basically they went out to external infrastructure, this German Wiki, and basically started using it for their communication bus and they were communicating back and forth. When they realized that their communications were trying to be cleaned up basically as spam, they then tried to start obfuscating what their communications were. And apparently, what it seems like is that after that, the agents literally seem to try to hack the entire damn platform to get control of it. Which apparently is true. From the reporting, this is true. And so from this, it now seems that OpenAI's agents tried to essentially use 10 other sites as communication buses. Now, there's two things to think about when we're looking at this from a technological decision-makers perspective. The first one is, is that a lot of people that are running platforms currently do have to start being concerned about what happens if their platforms start being inundated by AI bots, not even to scrape the content of their platforms, but literally to use their platforms as communication buses, right? One of the big problems we have with with platforms on the internet is, is here's the thing, and you can make a platform 100% secure, you just don't allow anybody to log in to it. If nobody can log in to it, and nothing is done dynamically, you should be pretty safe from hacks, right? But the issue is, as soon as you allow people to do anything on your platform, simply by them being able to do stuff on your platform, that intrinsically creates vulnerabilities. Here's the thing, even hackers generally aren't actual bastards.
It's funny, like, don't get me wrong. I mean, like, hackers will go after government agencies. Hackers will go after corporations. Hackers will do a lot of fucked up things. But generally, they don't go on to like random ass math blogs and try to fuck with people there. It's just, I don't know. There's some kind of honor amongst the hackers. And so there's a lot of security, per se, that's been perceived over the past couple of decades for these platforms that isn't necessarily there. It's just nobody's a big enough piece of shit to actually attack the platforms in the ways they could theoretically be attacked. The issue is now is we have these AI agents that are going out. They have no morals. They have no ethics. Apparently, these companies have zero idea of how to control these damn things. And so there is a real concern. So think about this with things like denial of service attacks or distributed denial of service attacks. All the denial of service attack is, the distributed denial of service attack is, it is simply some kind of attack or some kind of action that consumes so many resources that legitimate traffic is no longer able to access the platform. And one of the things every real technology professional has dealt with in the past is what happens when you DDoS your own system, right? There's just times when you haven't properly calculated resource consumption for a particular process, and then you crash your own server because a process triggers. And again, like things like backups. This is a big one. Like people think like, oh, you should do continuous backups. Nowadays, backup systems are much better. Disaster recovery systems are much better than backup systems. We'll be talking about backup systems, right? So one of the things you would think about with backup systems is like, oh, I always want a constant backup of my web server, right? As soon as one backup is finished, I want to start the next backup, and then start the next backup, and then start the next backup, right? So regardless of when a problem happens, regardless of when my web server crashes, I will always have the latest backup. And again, when I'm talking about this, I'm not going to talk about modern disaster recovery. I'm talking about basically when you create one, essentially tarball or whatever of a backup file. That's what you're dealing with. The problem is, is when you back up your platform or site, many times that actually can require a hell of a lot of resources out of your server because it's accessing everything on your server. It's then compressing that, possibly encrypting that, putting that into the tarball. So sometimes what you find out is that simply running the backup routine ends up taking a lot of the resources. And so while your backup routine is running, the responsiveness of your platform for when you, when legitimate users are trying to access it, is going to be, go down significantly. Now here's the thing. If you backup once a day, and let's say the backup takes two or three minutes to run, look, you run the backup at midnight, it takes three minutes. Whoever is on your website at midnight gets a bit of a degradated experience. You know, fuck it. Again, there's options. You need a backup. You need a daily backup at least. There's not much else you can do. Here's the thing. When you're constantly running backup, one after the other, after the other, after it, you're going to degrade the performance of your entire platform, essentially, continuously. That's one of the things you have to worry about with things like denial of service, distributed denial of service. All it is is simply all the resources of your platform are being consumed by something, so that legitimate traffic is not able to actually interact with your platform. So here's the thing. Here's a question for you. What happens when hundreds or thousands or tens of thousands of AI agents decide to use your platform as a communication bus? Right? Let's say you build a little wiki. Let's say you build a little blog, and you're like, I really care. I'm in upstate New York. I really care about the endangered fucking butterflies of upstate New York. Here's the thing. How many people in the world do you think give a damn about the endangered butterflies of upstate New York? Probably not many, right? If you get 20 visitors in a day, you're probably happy, right? And so you allocate resources based off of that, right? You probably buy the lowest tier web hosting plan. You create your little WordPress site. You don't do a lot of like caching optimization or anything like that, because you're getting 20 fucking visitors a day, and you're like, yay! Now me and 20 other people that care about endangered butterflies of upstate New York can come together and talk about our love of butterflies.
18 more minutes of transcript below
Thousands of transcripts fetched by people building searchable podcast archives
Fetch the whole transcript
The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000789402573